Mipmapped High Resolution Video for Immersive

1
Mipmapped High Resolution Video for Immersive
Virtual 3D Environments
Andrew Prudhomme
Advisor Jurgen Schulze
Simplified Plug-in Structure
Introduction
Video Frames
VideoLOD Node
The goal of this project was to create an
efficient way of displaying 2D high resolution
video in immersive 3D environments. An aspect of
this goal was the implementation of mipmapping on
the video stream. Since the stream can be moved
towards or away from the viewer and oriented for
different degrees of visibility, the area in
pixels of the video in the plane of the screen
was used to determine what level of detail was
required. This project took the form of a C
COVISE plug-in. The data sets were sequences of
prerecorded and preprocessed frames of 2k and 4k
resolutions. Another important aspect of the
project was to create a solution that scaled well
to an environment where one logical display is
being run by multiple computers, such as a tiled
display wall.
Each frame of the video is pre-encoded for use
with the plug-in. The image is run through
several tools that generate the frames mipmap
levels. In addition, the images are broken up
into multiple tiles. Each of these tiles are
saved into a single archives as a binary
representation of a scene graph node structure.
SLOD Node
SLOD Node
Image Tile
Image Tile
Group
Group
Image Tile
Image Tile
Image Tile
Image Tile
Image composed of eight tiles
Results
Mipmapping
Image Tile Visibility
The testing for this plug-in was performed on
both a single and multi node system. The testing
system was a cluster of 16 computers consisting
of a single master node and the rest driving a
tiled display wall. In this wall, the nodes were
arranged in three columns with five nodes each.
Each node drove four displays arranged
horizontally. The wall was used for multi node
testing and the masternode alone was used for
single node testing.
Determining which tiles are visible to the
display allows the program to selectively load
only the image data that is needed. The
visibility calculation use the same pixel
coordinates used to determine the mipmap level.
There are three possible conditions for a visible
tile.
Mipmapping involves using a lower quality version
of an image when possible.
The data sets used for test were selected to be
long enough to eliminate the computers' ability
to cache the entire stream. The base data set
was a set of 3000 4k video frames. From these
frames, four video sets were created and
processed as seen on the chart.
The level of detail needed for each image tile is
determined by the pixel area of the tile in the
plane of the screen. The corner coordinates of
each tile are translated into pixel coordinates
on screen through a series of matrix
multiplications. The area is found by bounding
the tile with a rectangle and subtracting the
extra pieces.
A tile corner is in the display area.
Performance on Tiled Display Wall
Data Set FPS Max Tiles Seen by a Single Node
HD-RGB24 3.3 6
HD-DXT1 11.0 6
4k-RGB24 2.4 8
4k-DXT1 6.5 8
A tile edge passes through the display area.
A tile entirely encompasses the display area.
The ratio of the screen pixel area to the actual
pixel area determines the level of detail
required for the tile.